Country / Region
EMEA
Tags
Artificial intelligence, Data quality, Research
High-quality clinical terminologies are essential for a connected and efficient healthcare ecosystem. Previous studies have shown the impact of quality issues on downstream applications, such as reduced recall and precision in cohort queries over EHRs. The literature presents several proposals to detect and/or resolve these quality issues. While these proposals typically employ either a lexical, structural, or machine learning-based approach, some combine different techniques to address the issues. This presentation outlines a series of experiments we conducted to improve the quality and clinical accuracy of the SNOMED CT terminology by identifying key areas for enhancement and proposing AI-enabled methods to automate the detection and correction of structural anomalies, such as primitive, misaligned, missing, and redundant concepts.
Description
The experiments we carried out aimed to address structural anomalies, such as primitive, misaligned, missing, and redundant concepts. Primitive concepts are those not fully defined by necessary conditions, making it impossible to automatically classify them or their subtypes into the hierarchy unless a sufficient condition exists for that concept. Intermediate primitives serve as both parents and children within the hierarchy, posing a challenge due to the manual effort they require. Misaligned concepts arise when the modelling of concepts does not follow the template for that sub-hierarchy, likely indicating inconsistent modelling within a sub-hierarchy. The missing concept anomaly refers to gaps where certain concepts that should logically exist within the hierarchy are absent, hindering accurate data representation and retrieval. The redundant concept anomaly occurs when multiple concepts essentially represent the same clinical idea, causing confusion among users, increasing clinical data fragmentation, and complicating the aggregation and analysis of information, along with the extra effort needed to maintain and update the terminology. Our experiments explored AI-enabled approaches, combined with lexical and logical-based techniques, to enhance the quality and clinical accuracy of the SNOMED CT terminology by identifying key areas for improvement and methods to automate the detection and correction of structural anomalies. The presentation will detail these approaches, the methods applied, the tools used, the outcomes achieved, and the lessons learned from the experimentation. A guideline for applying the more promising methods in real-world terminology authoring will also be outlined to conclude.
Scope
SNOMED CT is a widely used, multilingual clinical healthcare terminology that standardizes medical terms in electronic health records (EHRs). It enables healthcare providers to accurately document patient conditions, treatments, and outcomes using standardized terms, ensuring consistent recording and understanding of clinical information across various systems and locations. This standardization supports the seamless exchange of health information between healthcare providers and systems, facilitating coordinated patient care and public health reporting. Researchers also benefit, as it allows them to query and analyse large datasets of clinical information, leading to more reliable research outcomes through precise cohort identification and data analysis. Meanwhile, previous studies have shown the impact of quality issues on downstream applications. In conclusion, SNOMED CT is the perfect candidate for evaluating new techniques to address quality issues, given its comprehensiveness and crucial role in healthcare.
How SNOMED CT will be used
The SNOMED CT clinical terminology was chosen as the focus of our experiments aimed at assessing AI-driven methods to enhance its quality and clinical accuracy by automating the detection and correction of structural anomalies.
Why SNOMED CT will be used
Contact
